India promised a light touch on AI. Then came the three-hour rule.
Six months after New Delhi hosted the largest AI summit ever convened, the distance between India's innovation-first rhetoric and its hardest-edged internet rules has become the defining question in Indian technology policy.
In November 2025, the Ministry of Electronics and Information Technology released a document that told the world what kind of AI regulator India intended to be. The India AI Governance Guidelines ran on seven principles, six pillars and a preference — stated plainly — for voluntary compliance, self-certification and regulatory sandboxes over binding mandates. India would not pass an AI law. It would lean on the statutes it already had and get out of the way of its developers.
Three months later, on 10 February 2026, the same ministry notified an amendment to the Information Technology Rules giving online platforms three hours to act on certain government takedown directions, down from thirty-six. It is by some distance the shortest compliance window any government has imposed on internet intermediaries anywhere. The rules took effect on 20 February, ten days after notification.
Between those two documents sits the central tension of Indian AI policy, and six months on it has not resolved. India has built a permissive environment for people who build AI, and a markedly stricter one for the platforms where its output is seen.
The adoption case for moving fast
The reluctance to regulate models is not drift. It follows from a bet placed in March 2024, when the Cabinet cleared the IndiaAI Mission with a five-year corpus of ₹10,371.92 crore — roughly $1.25 billion — weighted heavily toward subsidised compute.
That bet has paid off on its own terms. The mission’s original target was 10,000 GPUs. It has deployed some 38,000, and on 17 February 2026 Union Minister Ashwini Vaishnaw announced a further 20,000, with the government targeting 100,000 by year-end.
Demand has kept pace. By most third-party estimates India is now either the largest or second-largest national market for consumer AI assistants, with user numbers in the tens of millions. Surveys of workplace AI use put India among a small group of countries where a substantial majority of employees report using these tools regularly.
The ministry’s argument is not weak: a country adopting a technology this quickly has more to lose from premature regulation than from delayed regulation.
The diplomacy
In February, India converted that position into standing. The India AI Impact Summit ran from 16 to 21 February at Bharat Mandapam in New Delhi, drawing delegations from 118 countries, more than twenty heads of government, and over a hundred chief executives — the fourth in the series that began at Bletchley Park and, by attendance, comfortably the largest.
Its output was the New Delhi Declaration on AI Impact, adopted on 19 February and endorsed by 89 countries and international organisations. Framed around the sutras of People, Planet and Progress, it committed signatories to the principle that AI’s benefits be shared equitably rather than concentrated in the few countries that build frontier models. Thirteen frontier developers signed parallel commitments on trustworthy deployment.
For a government that had declined to write an AI law, it was considerable: India had shaped the global conversation without regulating at home.
The hard edge
And then, in the same month, the IT Amendment Rules.
The amendment does something India’s intermediary framework had never done: it defines synthetically generated information as a distinct category and brings deepfakes inside the platform liability regime. Platforms must deploy technical measures against unlawful synthetic media, attach prominent labels and persistent provenance metadata to AI-generated content, and display that label continuously throughout visual content rather than only at its start.
The three-hour takedown window is the sharpest edge. Compressed from thirty-six hours, it leaves little room for the legal review platforms use to assess whether a direction is lawful. Failure puts safe harbour at risk — the protection that makes large-scale user-generated content viable in India at all.
Digital rights groups argue the combination is dangerous in a predictable way: a three-hour clock backed by loss of immunity gives platforms a structural incentive to remove first and assess later. Amnesty International called in April for the amendment’s withdrawal. Others argue the rules sit uneasily with Article 19 of the International Covenant on Civil and Political Rights, which requires restrictions on expression to satisfy legality, necessity and proportionality — and that a uniform three-hour window, applied regardless of the complexity of the content, struggles on proportionality.
The government’s position is that deepfakes move faster than thirty-six hours. On the narrow question of velocity, that is difficult to dispute.
What is actually unresolved
Reading this as hypocrisy is too easy. The two instruments regulate different things: the guidelines address who trains and deploys models, the IT Rules who distributes their output. A government can coherently be permissive about the first and strict about the second.
But the guidelines promised more than light-touch model regulation. They promised a style — voluntary, consultative, slow to mandate. The IT amendment was notified with ten days’ notice and imposed the world’s tightest takedown clock. Those are not the same instincts, and the gap between them is the real question: which document describes how India will govern AI over the next five years?
Three institutions recommended by the guidelines — an AI Governance Group, a Technology and Policy Expert Committee, and an AI Safety Institute — are the mechanism through which that gets answered. Their composition, their independence, and whether they are consulted before the next amendment rather than after are worth watching more closely than any further statement of principles.
Sources: Ministry of Electronics and Information Technology; Press Information Bureau; Ministry of External Affairs. Figures on AI adoption are drawn from third-party market estimates and vary by methodology.